maybe briefly explain why we went for a static method here

In .validate() of VideoNdArray we want to call the validation for general VideoNdArray (by calling super().validate()), and secondly we want to add a video specific shape check (and respectively for torch). As far as I know it's not possible to specify in super() which parent class to go to, instead it just goes to through all the parent classes and stops when it find a corresponding method. Therefore we have to call the second check explicitly. When calling it as as VideoTensorMixin.validate() the information of the child class gets lost though, which we need to get the corresponding comp backend. Therefore the changes in the signature and to staticmethod.

If you wan to use a specific superclass implementation, instead of using super() you can do something like:

SuperclassThatYouWant.foo(*inputs)

Or if you need to pass the subclass cls:

SuperclassThatYouWant.foo.__func__(cls, *inputs)

This unbinds foo from the class, so you can give it any cls argument that you want.

You can see this in the __class_getitem__ implementation of NdArray.

oh nice, thanks! This works, but mypy complains:

docarray/typing/tensor/video/video_ndarray.py:34: error: "Callable[[VideoTensorMixin], VideoTensorMixin]" has no attribute "__func__"  [attr-defined]

Do u know what to do here to satisfy mypy?

Ah ok just saw that in __class_getitem__ of NdArray it is being ignored ( # type: ignore), will add that here too then

why do we need to ignore the type ?

Without it mypy complains, I think because value in validate_shape can be an instance of VideoTensor, in our use case either VideoNdArray or VideoTorchTensor but the .n_dim methods have a specified input class respectiveley to their comp backend (np.ndarray or torch.Tensor). Therefore mypy complains that value is not an NdArray for NumpyCompBackend.n_dim() (or torch tensor for torch comp backend):

docarray/typing/tensor/video/video_tensor_mixin.py:22: error: Argument 1 to "n_dim" of "TorchCompBackend" has incompatible type "T"; expected "Tensor"  [arg-type]
docarray/typing/tensor/video/video_tensor_mixin.py:22: error: Argument 1 to "n_dim" of "NumpyCompBackend" has incompatible type "T"; expected "ndarray[Any, Any]"  [arg-type]

I think this is this issue that has to be fixed using the ugly overloads.

@JohannesMessner need ur help there we did not manage to debug this. I don't understand why it is not working and why we have different TypeVar in AbstractCompBackend

overload was not working here, because the Mixin should not know about the classes VideoNdArray and VideoTorchTensor

why not None as a default value ?

The .load() from VideoUrl returns an AudioNdArray (and some other stuff), which can't be written to video.audio.tensor if video.audio == None.

okay makes sense